{"id":"W2969250926","doi":"10.1080/0142159x.2019.1652260","title":"Are rating scales really better than checklists for measuring increasing levels of expertise?","year":2019,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":20,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"University of Ottawa","keywords":"Checklist; Rating scale; Psychology; Test (biology); Measure (data warehouse); Objective structured clinical examination; Medical education; Rating system; Applied psychology; Medicine; Clinical psychology; Computer science; Data mining; Cognitive psychology; Developmental psychology","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001920825,0.0001721678,0.0006126277,0.00005631998,0.00004319624,0.00001311141,0.0001489319,0.0002833462,0.001057309],"category_scores_gemma":[0.1128755,0.0001309307,0.0002209994,0.00009507133,0.0001343272,0.00004308216,0.00007237599,0.0003655052,0.00003650667],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006168526,"about_ca_system_score_gemma":0.000123569,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001081279,"about_ca_topic_score_gemma":0.00002093241,"domain_scores_codex":[0.9977852,0.0001260171,0.0005744087,0.000350212,0.0008146662,0.0003495273],"domain_scores_gemma":[0.9955441,0.003319424,0.0002460928,0.0003491306,0.0001375699,0.0004036847],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002333221,0.0003320063,0.9445738,0.0001807544,0.00008984844,0.00003383043,0.0004116876,9.08075e-7,0.002009251,0.00003563159,0.001371211,0.05072778],"study_design_scores_gemma":[0.004992517,0.0002710085,0.9769023,0.01003733,0.0001181104,0.00006948859,0.0003775145,0.003286993,0.002656011,0.0002118949,0.0007918962,0.0002849564],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9915117,0.0003254192,0.0007278654,0.003244041,0.0002216082,0.0003904917,0.000005524073,0.00007025879,0.003503081],"genre_scores_gemma":[0.9950989,0.000008571979,0.001952895,0.001072459,0.0005974965,0.00003056659,0.00001209125,0.00003988164,0.001187061],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1109547,"threshold_uncertainty_score":0.9998559,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06439051680890505,"score_gpt":0.3509525032084154,"score_spread":0.2865619863995103,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}